Image analysis for cervical neoplasia detection and diagnosis

Abstract

The present invention is an automated image analysis framework for cervical cancerous lesion detection. The present invention uses domain-specific diagnostic features in a probabilistic manner using conditional random fields. In addition, the present invention discloses a novel window-based performance assessment scheme for two-dimensional image analysis, which addresses the intrinsic problem of image misalignment. As a domain-specific anatomical feature, image regions corresponding to different tissue types are extracted from cervical images taken before and after the application of acetic acid during a clinical exam. The unique optical properties of each tissue type and the diagnostic relationships between neighboring regions are incorporated in the conditional random field model. The output provides information about both the tissue severity and the location of cancerous tissue in an image.

What is claimed is: 
     
       1. A process for detection and diagnosis of cancer in tissues, comprising:
 acquiring polarized and non-polarized images of said tissues; 
 normalizing said images to account for color and spatial variations; 
 registering said images to correct for tissue deformation; 
 generating an anatomical features map from said images using color and texture information; 
 identifying regions in said images of different tissue types based on said anatomical features map, whereby each region has only one tissue type; 
 segmenting sub-regions within each region that are homogeneous in color and intensity; 
 extracting diagnostically relevant features from each of said sub-regions, wherein said diagnostically relevant features are selected from the group consisting of acetowhitening and abnormal blood vessel features; and 
 classifying said sub-regions as normal or abnormal based on said extracted diagnostically relevant features in said sub-regions and probabilistic dependencies based on classification of neighboring regions; 
 wherein said extracting diagnostically relevant features step for acetowhitening is performed by calculating mean, standard deviation, entropy and ratios between different color channels of said images of said tissue. 
 
     
     
       2. A process according to  claim 1 , wherein said extracting diagnostically relevant features step is performed for abnormal vessel features by applying a linear rotating structuring element and a morphological transformation to automatically detect mosaicism, punctation, and atypical vessel patterns by extracting intercapillary distance between vessels, size of each vessel, and density of vessels. 
     
     
       3. A process according to  claim 1 , further comprising: automatically classifying tissues in each sub-region as normal or abnormal using a conditional random field classifier that incorporates probabilistic dependencies on classification of neighboring sub-regions. 
     
     
       4. A process according to  claim 1 , wherein said tissues are cervical tissues and said identifying step identifies said tissue types as squamous epithelium, columnar epithelium, cervical os, and transformation zone. 
     
     
       5. A process according to  claim 1 , wherein said images are red, green and blue visible light images. 
     
     
       6. A process according to  claim 1 , wherein said images are selected from the group consisting of pre-acetic acid images, post-acetic acid images, time course acetowhite images and reflectance and fluorescence images.